1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. 5 */ 6 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8 9 #include <linux/sched.h> 10 #include <linux/slab.h> 11 #include <linux/spinlock.h> 12 #include <linux/completion.h> 13 #include <linux/buffer_head.h> 14 #include <linux/blkdev.h> 15 #include <linux/kthread.h> 16 #include <linux/export.h> 17 #include <linux/namei.h> 18 #include <linux/mount.h> 19 #include <linux/gfs2_ondisk.h> 20 #include <linux/quotaops.h> 21 #include <linux/lockdep.h> 22 #include <linux/module.h> 23 #include <linux/backing-dev.h> 24 #include <linux/fs_parser.h> 25 26 #include "gfs2.h" 27 #include "incore.h" 28 #include "bmap.h" 29 #include "glock.h" 30 #include "glops.h" 31 #include "inode.h" 32 #include "recovery.h" 33 #include "rgrp.h" 34 #include "super.h" 35 #include "sys.h" 36 #include "util.h" 37 #include "log.h" 38 #include "quota.h" 39 #include "dir.h" 40 #include "meta_io.h" 41 #include "trace_gfs2.h" 42 #include "lops.h" 43 44 #define DO 0 45 #define UNDO 1 46 47 /** 48 * gfs2_tune_init - Fill a gfs2_tune structure with default values 49 * @gt: tune 50 * 51 */ 52 53 static void gfs2_tune_init(struct gfs2_tune *gt) 54 { 55 spin_lock_init(>->gt_spin); 56 57 gt->gt_quota_warn_period = 10; 58 gt->gt_quota_scale_num = 1; 59 gt->gt_quota_scale_den = 1; 60 gt->gt_new_files_jdata = 0; 61 gt->gt_max_readahead = BIT(18); 62 gt->gt_complain_secs = 10; 63 } 64 65 void free_sbd(struct gfs2_sbd *sdp) 66 { 67 if (sdp->sd_lkstats) 68 free_percpu(sdp->sd_lkstats); 69 kfree(sdp); 70 } 71 72 static struct gfs2_sbd *init_sbd(struct super_block *sb) 73 { 74 struct gfs2_sbd *sdp; 75 struct address_space *mapping; 76 77 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL); 78 if (!sdp) 79 return NULL; 80 81 sdp->sd_vfs = sb; 82 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats); 83 if (!sdp->sd_lkstats) 84 goto fail; 85 sb->s_fs_info = sdp; 86 87 set_bit(SDF_NOJOURNALID, &sdp->sd_flags); 88 gfs2_tune_init(&sdp->sd_tune); 89 90 init_waitqueue_head(&sdp->sd_glock_wait); 91 init_waitqueue_head(&sdp->sd_async_glock_wait); 92 atomic_set(&sdp->sd_glock_disposal, 0); 93 init_completion(&sdp->sd_locking_init); 94 init_completion(&sdp->sd_wdack); 95 spin_lock_init(&sdp->sd_statfs_spin); 96 97 spin_lock_init(&sdp->sd_rindex_spin); 98 sdp->sd_rindex_tree.rb_node = NULL; 99 100 INIT_LIST_HEAD(&sdp->sd_jindex_list); 101 spin_lock_init(&sdp->sd_jindex_spin); 102 mutex_init(&sdp->sd_jindex_mutex); 103 init_completion(&sdp->sd_journal_ready); 104 105 INIT_LIST_HEAD(&sdp->sd_quota_list); 106 mutex_init(&sdp->sd_quota_mutex); 107 mutex_init(&sdp->sd_quota_sync_mutex); 108 init_waitqueue_head(&sdp->sd_quota_wait); 109 INIT_LIST_HEAD(&sdp->sd_trunc_list); 110 spin_lock_init(&sdp->sd_trunc_lock); 111 spin_lock_init(&sdp->sd_bitmap_lock); 112 113 INIT_LIST_HEAD(&sdp->sd_sc_inodes_list); 114 115 mapping = &sdp->sd_aspace; 116 117 address_space_init_once(mapping); 118 mapping->a_ops = &gfs2_rgrp_aops; 119 mapping->host = sb->s_bdev->bd_inode; 120 mapping->flags = 0; 121 mapping_set_gfp_mask(mapping, GFP_NOFS); 122 mapping->private_data = NULL; 123 mapping->writeback_index = 0; 124 125 spin_lock_init(&sdp->sd_log_lock); 126 atomic_set(&sdp->sd_log_pinned, 0); 127 INIT_LIST_HEAD(&sdp->sd_log_revokes); 128 INIT_LIST_HEAD(&sdp->sd_log_ordered); 129 spin_lock_init(&sdp->sd_ordered_lock); 130 131 init_waitqueue_head(&sdp->sd_log_waitq); 132 init_waitqueue_head(&sdp->sd_logd_waitq); 133 spin_lock_init(&sdp->sd_ail_lock); 134 INIT_LIST_HEAD(&sdp->sd_ail1_list); 135 INIT_LIST_HEAD(&sdp->sd_ail2_list); 136 137 init_rwsem(&sdp->sd_log_flush_lock); 138 atomic_set(&sdp->sd_log_in_flight, 0); 139 init_waitqueue_head(&sdp->sd_log_flush_wait); 140 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); 141 mutex_init(&sdp->sd_freeze_mutex); 142 143 return sdp; 144 145 fail: 146 free_sbd(sdp); 147 return NULL; 148 } 149 150 /** 151 * gfs2_check_sb - Check superblock 152 * @sdp: the filesystem 153 * @sb: The superblock 154 * @silent: Don't print a message if the check fails 155 * 156 * Checks the version code of the FS is one that we understand how to 157 * read and that the sizes of the various on-disk structures have not 158 * changed. 159 */ 160 161 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent) 162 { 163 struct gfs2_sb_host *sb = &sdp->sd_sb; 164 165 if (sb->sb_magic != GFS2_MAGIC || 166 sb->sb_type != GFS2_METATYPE_SB) { 167 if (!silent) 168 pr_warn("not a GFS2 filesystem\n"); 169 return -EINVAL; 170 } 171 172 if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN || 173 sb->sb_fs_format > GFS2_FS_FORMAT_MAX || 174 sb->sb_multihost_format != GFS2_FORMAT_MULTI) { 175 fs_warn(sdp, "Unknown on-disk format, unable to mount\n"); 176 return -EINVAL; 177 } 178 179 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE || 180 (sb->sb_bsize & (sb->sb_bsize - 1))) { 181 pr_warn("Invalid block size\n"); 182 return -EINVAL; 183 } 184 185 return 0; 186 } 187 188 static void end_bio_io_page(struct bio *bio) 189 { 190 struct page *page = bio->bi_private; 191 192 if (!bio->bi_status) 193 SetPageUptodate(page); 194 else 195 pr_warn("error %d reading superblock\n", bio->bi_status); 196 unlock_page(page); 197 } 198 199 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf) 200 { 201 struct gfs2_sb_host *sb = &sdp->sd_sb; 202 struct super_block *s = sdp->sd_vfs; 203 const struct gfs2_sb *str = buf; 204 205 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic); 206 sb->sb_type = be32_to_cpu(str->sb_header.mh_type); 207 sb->sb_format = be32_to_cpu(str->sb_header.mh_format); 208 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format); 209 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format); 210 sb->sb_bsize = be32_to_cpu(str->sb_bsize); 211 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift); 212 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr); 213 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino); 214 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr); 215 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino); 216 217 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN); 218 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN); 219 memcpy(&s->s_uuid, str->sb_uuid, 16); 220 } 221 222 /** 223 * gfs2_read_super - Read the gfs2 super block from disk 224 * @sdp: The GFS2 super block 225 * @sector: The location of the super block 226 * @error: The error code to return 227 * 228 * This uses the bio functions to read the super block from disk 229 * because we want to be 100% sure that we never read cached data. 230 * A super block is read twice only during each GFS2 mount and is 231 * never written to by the filesystem. The first time its read no 232 * locks are held, and the only details which are looked at are those 233 * relating to the locking protocol. Once locking is up and working, 234 * the sb is read again under the lock to establish the location of 235 * the master directory (contains pointers to journals etc) and the 236 * root directory. 237 * 238 * Returns: 0 on success or error 239 */ 240 241 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent) 242 { 243 struct super_block *sb = sdp->sd_vfs; 244 struct gfs2_sb *p; 245 struct page *page; 246 struct bio *bio; 247 248 page = alloc_page(GFP_NOFS); 249 if (unlikely(!page)) 250 return -ENOMEM; 251 252 ClearPageUptodate(page); 253 ClearPageDirty(page); 254 lock_page(page); 255 256 bio = bio_alloc(GFP_NOFS, 1); 257 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9); 258 bio_set_dev(bio, sb->s_bdev); 259 bio_add_page(bio, page, PAGE_SIZE, 0); 260 261 bio->bi_end_io = end_bio_io_page; 262 bio->bi_private = page; 263 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META); 264 submit_bio(bio); 265 wait_on_page_locked(page); 266 bio_put(bio); 267 if (!PageUptodate(page)) { 268 __free_page(page); 269 return -EIO; 270 } 271 p = kmap(page); 272 gfs2_sb_in(sdp, p); 273 kunmap(page); 274 __free_page(page); 275 return gfs2_check_sb(sdp, silent); 276 } 277 278 /** 279 * gfs2_read_sb - Read super block 280 * @sdp: The GFS2 superblock 281 * @silent: Don't print message if mount fails 282 * 283 */ 284 285 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent) 286 { 287 u32 hash_blocks, ind_blocks, leaf_blocks; 288 u32 tmp_blocks; 289 unsigned int x; 290 int error; 291 292 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 293 if (error) { 294 if (!silent) 295 fs_err(sdp, "can't read superblock\n"); 296 return error; 297 } 298 299 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 300 GFS2_BASIC_BLOCK_SHIFT; 301 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 302 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize - 303 sizeof(struct gfs2_dinode)) / sizeof(u64); 304 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize - 305 sizeof(struct gfs2_meta_header)) / sizeof(u64); 306 sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize - 307 sizeof(struct gfs2_log_descriptor)) / sizeof(u64); 308 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header); 309 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2; 310 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1; 311 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64); 312 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize - 313 sizeof(struct gfs2_meta_header)) / 314 sizeof(struct gfs2_quota_change); 315 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize - 316 sizeof(struct gfs2_meta_header)) 317 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */ 318 319 /* 320 * We always keep at least one block reserved for revokes in 321 * transactions. This greatly simplifies allocating additional 322 * revoke blocks. 323 */ 324 atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs); 325 326 /* Compute maximum reservation required to add a entry to a directory */ 327 328 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH), 329 sdp->sd_jbsize); 330 331 ind_blocks = 0; 332 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) { 333 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs); 334 ind_blocks += tmp_blocks; 335 } 336 337 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH; 338 339 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks; 340 341 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize - 342 sizeof(struct gfs2_dinode); 343 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs; 344 for (x = 2;; x++) { 345 u64 space, d; 346 u32 m; 347 348 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs; 349 d = space; 350 m = do_div(d, sdp->sd_inptrs); 351 352 if (d != sdp->sd_heightsize[x - 1] || m) 353 break; 354 sdp->sd_heightsize[x] = space; 355 } 356 sdp->sd_max_height = x; 357 sdp->sd_heightsize[x] = ~0; 358 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT); 359 360 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize - 361 sizeof(struct gfs2_leaf)) / 362 GFS2_MIN_DIRENT_SIZE; 363 return 0; 364 } 365 366 static int init_names(struct gfs2_sbd *sdp, int silent) 367 { 368 char *proto, *table; 369 int error = 0; 370 371 proto = sdp->sd_args.ar_lockproto; 372 table = sdp->sd_args.ar_locktable; 373 374 /* Try to autodetect */ 375 376 if (!proto[0] || !table[0]) { 377 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 378 if (error) 379 return error; 380 381 if (!proto[0]) 382 proto = sdp->sd_sb.sb_lockproto; 383 if (!table[0]) 384 table = sdp->sd_sb.sb_locktable; 385 } 386 387 if (!table[0]) 388 table = sdp->sd_vfs->s_id; 389 390 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN); 391 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN); 392 393 table = sdp->sd_table_name; 394 while ((table = strchr(table, '/'))) 395 *table = '_'; 396 397 return error; 398 } 399 400 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh, 401 int undo) 402 { 403 int error = 0; 404 405 if (undo) 406 goto fail_trans; 407 408 error = gfs2_glock_nq_num(sdp, 409 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops, 410 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE, 411 mount_gh); 412 if (error) { 413 fs_err(sdp, "can't acquire mount glock: %d\n", error); 414 goto fail; 415 } 416 417 error = gfs2_glock_nq_num(sdp, 418 GFS2_LIVE_LOCK, &gfs2_nondisk_glops, 419 LM_ST_SHARED, 420 LM_FLAG_NOEXP | GL_EXACT, 421 &sdp->sd_live_gh); 422 if (error) { 423 fs_err(sdp, "can't acquire live glock: %d\n", error); 424 goto fail_mount; 425 } 426 427 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops, 428 CREATE, &sdp->sd_rename_gl); 429 if (error) { 430 fs_err(sdp, "can't create rename glock: %d\n", error); 431 goto fail_live; 432 } 433 434 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops, 435 CREATE, &sdp->sd_freeze_gl); 436 if (error) { 437 fs_err(sdp, "can't create transaction glock: %d\n", error); 438 goto fail_rename; 439 } 440 441 return 0; 442 443 fail_trans: 444 gfs2_glock_put(sdp->sd_freeze_gl); 445 fail_rename: 446 gfs2_glock_put(sdp->sd_rename_gl); 447 fail_live: 448 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 449 fail_mount: 450 gfs2_glock_dq_uninit(mount_gh); 451 fail: 452 return error; 453 } 454 455 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr, 456 u64 no_addr, const char *name) 457 { 458 struct gfs2_sbd *sdp = sb->s_fs_info; 459 struct dentry *dentry; 460 struct inode *inode; 461 462 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 463 GFS2_BLKST_FREE /* ignore */); 464 if (IS_ERR(inode)) { 465 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode)); 466 return PTR_ERR(inode); 467 } 468 dentry = d_make_root(inode); 469 if (!dentry) { 470 fs_err(sdp, "can't alloc %s dentry\n", name); 471 return -ENOMEM; 472 } 473 *dptr = dentry; 474 return 0; 475 } 476 477 static int init_sb(struct gfs2_sbd *sdp, int silent) 478 { 479 struct super_block *sb = sdp->sd_vfs; 480 struct gfs2_holder sb_gh; 481 u64 no_addr; 482 int ret; 483 484 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops, 485 LM_ST_SHARED, 0, &sb_gh); 486 if (ret) { 487 fs_err(sdp, "can't acquire superblock glock: %d\n", ret); 488 return ret; 489 } 490 491 ret = gfs2_read_sb(sdp, silent); 492 if (ret) { 493 fs_err(sdp, "can't read superblock: %d\n", ret); 494 goto out; 495 } 496 497 switch(sdp->sd_sb.sb_fs_format) { 498 case GFS2_FS_FORMAT_MAX: 499 sb->s_xattr = gfs2_xattr_handlers_max; 500 break; 501 502 case GFS2_FS_FORMAT_MIN: 503 sb->s_xattr = gfs2_xattr_handlers_min; 504 break; 505 506 default: 507 BUG(); 508 } 509 510 /* Set up the buffer cache and SB for real */ 511 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) { 512 ret = -EINVAL; 513 fs_err(sdp, "FS block size (%u) is too small for device " 514 "block size (%u)\n", 515 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev)); 516 goto out; 517 } 518 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) { 519 ret = -EINVAL; 520 fs_err(sdp, "FS block size (%u) is too big for machine " 521 "page size (%u)\n", 522 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE); 523 goto out; 524 } 525 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize); 526 527 /* Get the root inode */ 528 no_addr = sdp->sd_sb.sb_root_dir.no_addr; 529 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root"); 530 if (ret) 531 goto out; 532 533 /* Get the master inode */ 534 no_addr = sdp->sd_sb.sb_master_dir.no_addr; 535 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master"); 536 if (ret) { 537 dput(sdp->sd_root_dir); 538 goto out; 539 } 540 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir); 541 out: 542 gfs2_glock_dq_uninit(&sb_gh); 543 return ret; 544 } 545 546 static void gfs2_others_may_mount(struct gfs2_sbd *sdp) 547 { 548 char *message = "FIRSTMOUNT=Done"; 549 char *envp[] = { message, NULL }; 550 551 fs_info(sdp, "first mount done, others may mount\n"); 552 553 if (sdp->sd_lockstruct.ls_ops->lm_first_done) 554 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp); 555 556 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp); 557 } 558 559 /** 560 * gfs2_jindex_hold - Grab a lock on the jindex 561 * @sdp: The GFS2 superblock 562 * @ji_gh: the holder for the jindex glock 563 * 564 * Returns: errno 565 */ 566 567 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh) 568 { 569 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex); 570 struct qstr name; 571 char buf[20]; 572 struct gfs2_jdesc *jd; 573 int error; 574 575 name.name = buf; 576 577 mutex_lock(&sdp->sd_jindex_mutex); 578 579 for (;;) { 580 struct gfs2_inode *jip; 581 582 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh); 583 if (error) 584 break; 585 586 name.len = sprintf(buf, "journal%u", sdp->sd_journals); 587 name.hash = gfs2_disk_hash(name.name, name.len); 588 589 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL); 590 if (error == -ENOENT) { 591 error = 0; 592 break; 593 } 594 595 gfs2_glock_dq_uninit(ji_gh); 596 597 if (error) 598 break; 599 600 error = -ENOMEM; 601 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL); 602 if (!jd) 603 break; 604 605 INIT_LIST_HEAD(&jd->extent_list); 606 INIT_LIST_HEAD(&jd->jd_revoke_list); 607 608 INIT_WORK(&jd->jd_work, gfs2_recover_func); 609 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1); 610 if (IS_ERR_OR_NULL(jd->jd_inode)) { 611 if (!jd->jd_inode) 612 error = -ENOENT; 613 else 614 error = PTR_ERR(jd->jd_inode); 615 kfree(jd); 616 break; 617 } 618 619 spin_lock(&sdp->sd_jindex_spin); 620 jd->jd_jid = sdp->sd_journals++; 621 jip = GFS2_I(jd->jd_inode); 622 jd->jd_no_addr = jip->i_no_addr; 623 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list); 624 spin_unlock(&sdp->sd_jindex_spin); 625 } 626 627 mutex_unlock(&sdp->sd_jindex_mutex); 628 629 return error; 630 } 631 632 /** 633 * init_statfs - look up and initialize master and local (per node) statfs inodes 634 * @sdp: The GFS2 superblock 635 * 636 * This should be called after the jindex is initialized in init_journal() and 637 * before gfs2_journal_recovery() is called because we need to be able to write 638 * to these inodes during recovery. 639 * 640 * Returns: errno 641 */ 642 static int init_statfs(struct gfs2_sbd *sdp) 643 { 644 int error = 0; 645 struct inode *master = d_inode(sdp->sd_master_dir); 646 struct inode *pn = NULL; 647 char buf[30]; 648 struct gfs2_jdesc *jd; 649 struct gfs2_inode *ip; 650 651 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs"); 652 if (IS_ERR(sdp->sd_statfs_inode)) { 653 error = PTR_ERR(sdp->sd_statfs_inode); 654 fs_err(sdp, "can't read in statfs inode: %d\n", error); 655 goto out; 656 } 657 if (sdp->sd_args.ar_spectator) 658 goto out; 659 660 pn = gfs2_lookup_simple(master, "per_node"); 661 if (IS_ERR(pn)) { 662 error = PTR_ERR(pn); 663 fs_err(sdp, "can't find per_node directory: %d\n", error); 664 goto put_statfs; 665 } 666 667 /* For each jid, lookup the corresponding local statfs inode in the 668 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */ 669 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 670 struct local_statfs_inode *lsi = 671 kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS); 672 if (!lsi) { 673 error = -ENOMEM; 674 goto free_local; 675 } 676 sprintf(buf, "statfs_change%u", jd->jd_jid); 677 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf); 678 if (IS_ERR(lsi->si_sc_inode)) { 679 error = PTR_ERR(lsi->si_sc_inode); 680 fs_err(sdp, "can't find local \"sc\" file#%u: %d\n", 681 jd->jd_jid, error); 682 goto free_local; 683 } 684 lsi->si_jid = jd->jd_jid; 685 if (jd->jd_jid == sdp->sd_jdesc->jd_jid) 686 sdp->sd_sc_inode = lsi->si_sc_inode; 687 688 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list); 689 } 690 691 iput(pn); 692 ip = GFS2_I(sdp->sd_sc_inode); 693 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 694 &sdp->sd_sc_gh); 695 if (error) { 696 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error); 697 goto free_local; 698 } 699 return 0; 700 701 free_local: 702 free_local_statfs_inodes(sdp); 703 iput(pn); 704 put_statfs: 705 iput(sdp->sd_statfs_inode); 706 out: 707 return error; 708 } 709 710 /* Uninitialize and free up memory used by the list of statfs inodes */ 711 static void uninit_statfs(struct gfs2_sbd *sdp) 712 { 713 if (!sdp->sd_args.ar_spectator) { 714 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 715 free_local_statfs_inodes(sdp); 716 } 717 iput(sdp->sd_statfs_inode); 718 } 719 720 static int init_journal(struct gfs2_sbd *sdp, int undo) 721 { 722 struct inode *master = d_inode(sdp->sd_master_dir); 723 struct gfs2_holder ji_gh; 724 struct gfs2_inode *ip; 725 int jindex = 1; 726 int error = 0; 727 728 if (undo) { 729 jindex = 0; 730 goto fail_statfs; 731 } 732 733 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex"); 734 if (IS_ERR(sdp->sd_jindex)) { 735 fs_err(sdp, "can't lookup journal index: %d\n", error); 736 return PTR_ERR(sdp->sd_jindex); 737 } 738 739 /* Load in the journal index special file */ 740 741 error = gfs2_jindex_hold(sdp, &ji_gh); 742 if (error) { 743 fs_err(sdp, "can't read journal index: %d\n", error); 744 goto fail; 745 } 746 747 error = -EUSERS; 748 if (!gfs2_jindex_size(sdp)) { 749 fs_err(sdp, "no journals!\n"); 750 goto fail_jindex; 751 } 752 753 atomic_set(&sdp->sd_log_blks_needed, 0); 754 if (sdp->sd_args.ar_spectator) { 755 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0); 756 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 757 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 758 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 759 } else { 760 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) { 761 fs_err(sdp, "can't mount journal #%u\n", 762 sdp->sd_lockstruct.ls_jid); 763 fs_err(sdp, "there are only %u journals (0 - %u)\n", 764 gfs2_jindex_size(sdp), 765 gfs2_jindex_size(sdp) - 1); 766 goto fail_jindex; 767 } 768 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid); 769 770 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid, 771 &gfs2_journal_glops, 772 LM_ST_EXCLUSIVE, 773 LM_FLAG_NOEXP | GL_NOCACHE, 774 &sdp->sd_journal_gh); 775 if (error) { 776 fs_err(sdp, "can't acquire journal glock: %d\n", error); 777 goto fail_jindex; 778 } 779 780 ip = GFS2_I(sdp->sd_jdesc->jd_inode); 781 sdp->sd_jinode_gl = ip->i_gl; 782 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 783 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE, 784 &sdp->sd_jinode_gh); 785 if (error) { 786 fs_err(sdp, "can't acquire journal inode glock: %d\n", 787 error); 788 goto fail_journal_gh; 789 } 790 791 error = gfs2_jdesc_check(sdp->sd_jdesc); 792 if (error) { 793 fs_err(sdp, "my journal (%u) is bad: %d\n", 794 sdp->sd_jdesc->jd_jid, error); 795 goto fail_jinode_gh; 796 } 797 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 798 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 799 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 800 801 /* Map the extents for this journal's blocks */ 802 gfs2_map_journal_extents(sdp, sdp->sd_jdesc); 803 } 804 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free)); 805 806 /* Lookup statfs inodes here so journal recovery can use them. */ 807 error = init_statfs(sdp); 808 if (error) 809 goto fail_jinode_gh; 810 811 if (sdp->sd_lockstruct.ls_first) { 812 unsigned int x; 813 for (x = 0; x < sdp->sd_journals; x++) { 814 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x); 815 816 if (sdp->sd_args.ar_spectator) { 817 error = check_journal_clean(sdp, jd, true); 818 if (error) 819 goto fail_statfs; 820 continue; 821 } 822 error = gfs2_recover_journal(jd, true); 823 if (error) { 824 fs_err(sdp, "error recovering journal %u: %d\n", 825 x, error); 826 goto fail_statfs; 827 } 828 } 829 830 gfs2_others_may_mount(sdp); 831 } else if (!sdp->sd_args.ar_spectator) { 832 error = gfs2_recover_journal(sdp->sd_jdesc, true); 833 if (error) { 834 fs_err(sdp, "error recovering my journal: %d\n", error); 835 goto fail_statfs; 836 } 837 } 838 839 sdp->sd_log_idle = 1; 840 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags); 841 gfs2_glock_dq_uninit(&ji_gh); 842 jindex = 0; 843 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func); 844 return 0; 845 846 fail_statfs: 847 uninit_statfs(sdp); 848 fail_jinode_gh: 849 /* A withdraw may have done dq/uninit so now we need to check it */ 850 if (!sdp->sd_args.ar_spectator && 851 gfs2_holder_initialized(&sdp->sd_jinode_gh)) 852 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 853 fail_journal_gh: 854 if (!sdp->sd_args.ar_spectator && 855 gfs2_holder_initialized(&sdp->sd_journal_gh)) 856 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 857 fail_jindex: 858 gfs2_jindex_free(sdp); 859 if (jindex) 860 gfs2_glock_dq_uninit(&ji_gh); 861 fail: 862 iput(sdp->sd_jindex); 863 return error; 864 } 865 866 static struct lock_class_key gfs2_quota_imutex_key; 867 868 static int init_inodes(struct gfs2_sbd *sdp, int undo) 869 { 870 int error = 0; 871 struct inode *master = d_inode(sdp->sd_master_dir); 872 873 if (undo) 874 goto fail_qinode; 875 876 error = init_journal(sdp, undo); 877 complete_all(&sdp->sd_journal_ready); 878 if (error) 879 goto fail; 880 881 /* Read in the resource index inode */ 882 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex"); 883 if (IS_ERR(sdp->sd_rindex)) { 884 error = PTR_ERR(sdp->sd_rindex); 885 fs_err(sdp, "can't get resource index inode: %d\n", error); 886 goto fail_journal; 887 } 888 sdp->sd_rindex_uptodate = 0; 889 890 /* Read in the quota inode */ 891 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota"); 892 if (IS_ERR(sdp->sd_quota_inode)) { 893 error = PTR_ERR(sdp->sd_quota_inode); 894 fs_err(sdp, "can't get quota file inode: %d\n", error); 895 goto fail_rindex; 896 } 897 /* 898 * i_rwsem on quota files is special. Since this inode is hidden system 899 * file, we are safe to define locking ourselves. 900 */ 901 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem, 902 &gfs2_quota_imutex_key); 903 904 error = gfs2_rindex_update(sdp); 905 if (error) 906 goto fail_qinode; 907 908 return 0; 909 910 fail_qinode: 911 iput(sdp->sd_quota_inode); 912 fail_rindex: 913 gfs2_clear_rgrpd(sdp); 914 iput(sdp->sd_rindex); 915 fail_journal: 916 init_journal(sdp, UNDO); 917 fail: 918 return error; 919 } 920 921 static int init_per_node(struct gfs2_sbd *sdp, int undo) 922 { 923 struct inode *pn = NULL; 924 char buf[30]; 925 int error = 0; 926 struct gfs2_inode *ip; 927 struct inode *master = d_inode(sdp->sd_master_dir); 928 929 if (sdp->sd_args.ar_spectator) 930 return 0; 931 932 if (undo) 933 goto fail_qc_gh; 934 935 pn = gfs2_lookup_simple(master, "per_node"); 936 if (IS_ERR(pn)) { 937 error = PTR_ERR(pn); 938 fs_err(sdp, "can't find per_node directory: %d\n", error); 939 return error; 940 } 941 942 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid); 943 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf); 944 if (IS_ERR(sdp->sd_qc_inode)) { 945 error = PTR_ERR(sdp->sd_qc_inode); 946 fs_err(sdp, "can't find local \"qc\" file: %d\n", error); 947 goto fail_ut_i; 948 } 949 950 iput(pn); 951 pn = NULL; 952 953 ip = GFS2_I(sdp->sd_qc_inode); 954 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 955 &sdp->sd_qc_gh); 956 if (error) { 957 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error); 958 goto fail_qc_i; 959 } 960 961 return 0; 962 963 fail_qc_gh: 964 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 965 fail_qc_i: 966 iput(sdp->sd_qc_inode); 967 fail_ut_i: 968 iput(pn); 969 return error; 970 } 971 972 static const match_table_t nolock_tokens = { 973 { Opt_jid, "jid=%d", }, 974 { Opt_err, NULL }, 975 }; 976 977 static const struct lm_lockops nolock_ops = { 978 .lm_proto_name = "lock_nolock", 979 .lm_put_lock = gfs2_glock_free, 980 .lm_tokens = &nolock_tokens, 981 }; 982 983 /** 984 * gfs2_lm_mount - mount a locking protocol 985 * @sdp: the filesystem 986 * @args: mount arguments 987 * @silent: if 1, don't complain if the FS isn't a GFS2 fs 988 * 989 * Returns: errno 990 */ 991 992 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent) 993 { 994 const struct lm_lockops *lm; 995 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 996 struct gfs2_args *args = &sdp->sd_args; 997 const char *proto = sdp->sd_proto_name; 998 const char *table = sdp->sd_table_name; 999 char *o, *options; 1000 int ret; 1001 1002 if (!strcmp("lock_nolock", proto)) { 1003 lm = &nolock_ops; 1004 sdp->sd_args.ar_localflocks = 1; 1005 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1006 } else if (!strcmp("lock_dlm", proto)) { 1007 lm = &gfs2_dlm_ops; 1008 #endif 1009 } else { 1010 pr_info("can't find protocol %s\n", proto); 1011 return -ENOENT; 1012 } 1013 1014 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table); 1015 1016 ls->ls_ops = lm; 1017 ls->ls_first = 1; 1018 1019 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) { 1020 substring_t tmp[MAX_OPT_ARGS]; 1021 int token, option; 1022 1023 if (!o || !*o) 1024 continue; 1025 1026 token = match_token(o, *lm->lm_tokens, tmp); 1027 switch (token) { 1028 case Opt_jid: 1029 ret = match_int(&tmp[0], &option); 1030 if (ret || option < 0) 1031 goto hostdata_error; 1032 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags)) 1033 ls->ls_jid = option; 1034 break; 1035 case Opt_id: 1036 case Opt_nodir: 1037 /* Obsolete, but left for backward compat purposes */ 1038 break; 1039 case Opt_first: 1040 ret = match_int(&tmp[0], &option); 1041 if (ret || (option != 0 && option != 1)) 1042 goto hostdata_error; 1043 ls->ls_first = option; 1044 break; 1045 case Opt_err: 1046 default: 1047 hostdata_error: 1048 fs_info(sdp, "unknown hostdata (%s)\n", o); 1049 return -EINVAL; 1050 } 1051 } 1052 1053 if (lm->lm_mount == NULL) { 1054 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format); 1055 complete_all(&sdp->sd_locking_init); 1056 return 0; 1057 } 1058 ret = lm->lm_mount(sdp, table); 1059 if (ret == 0) 1060 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n", 1061 sdp->sd_sb.sb_fs_format); 1062 complete_all(&sdp->sd_locking_init); 1063 return ret; 1064 } 1065 1066 void gfs2_lm_unmount(struct gfs2_sbd *sdp) 1067 { 1068 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops; 1069 if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount) 1070 lm->lm_unmount(sdp); 1071 } 1072 1073 static int wait_on_journal(struct gfs2_sbd *sdp) 1074 { 1075 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL) 1076 return 0; 1077 1078 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE) 1079 ? -EINTR : 0; 1080 } 1081 1082 void gfs2_online_uevent(struct gfs2_sbd *sdp) 1083 { 1084 struct super_block *sb = sdp->sd_vfs; 1085 char ro[20]; 1086 char spectator[20]; 1087 char *envp[] = { ro, spectator, NULL }; 1088 sprintf(ro, "RDONLY=%d", sb_rdonly(sb)); 1089 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0); 1090 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp); 1091 } 1092 1093 /** 1094 * gfs2_fill_super - Read in superblock 1095 * @sb: The VFS superblock 1096 * @args: Mount options 1097 * @silent: Don't complain if it's not a GFS2 filesystem 1098 * 1099 * Returns: -errno 1100 */ 1101 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc) 1102 { 1103 struct gfs2_args *args = fc->fs_private; 1104 int silent = fc->sb_flags & SB_SILENT; 1105 struct gfs2_sbd *sdp; 1106 struct gfs2_holder mount_gh; 1107 struct gfs2_holder freeze_gh; 1108 int error; 1109 1110 sdp = init_sbd(sb); 1111 if (!sdp) { 1112 pr_warn("can't alloc struct gfs2_sbd\n"); 1113 return -ENOMEM; 1114 } 1115 sdp->sd_args = *args; 1116 1117 if (sdp->sd_args.ar_spectator) { 1118 sb->s_flags |= SB_RDONLY; 1119 set_bit(SDF_RORECOVERY, &sdp->sd_flags); 1120 } 1121 if (sdp->sd_args.ar_posix_acl) 1122 sb->s_flags |= SB_POSIXACL; 1123 if (sdp->sd_args.ar_nobarrier) 1124 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1125 1126 sb->s_flags |= SB_NOSEC; 1127 sb->s_magic = GFS2_MAGIC; 1128 sb->s_op = &gfs2_super_ops; 1129 sb->s_d_op = &gfs2_dops; 1130 sb->s_export_op = &gfs2_export_ops; 1131 sb->s_qcop = &gfs2_quotactl_ops; 1132 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; 1133 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE; 1134 sb->s_time_gran = 1; 1135 sb->s_maxbytes = MAX_LFS_FILESIZE; 1136 1137 /* Set up the buffer cache and fill in some fake block size values 1138 to allow us to read-in the on-disk superblock. */ 1139 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK); 1140 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits; 1141 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 1142 GFS2_BASIC_BLOCK_SHIFT; 1143 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 1144 1145 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit; 1146 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum; 1147 if (sdp->sd_args.ar_statfs_quantum) { 1148 sdp->sd_tune.gt_statfs_slow = 0; 1149 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum; 1150 } else { 1151 sdp->sd_tune.gt_statfs_slow = 1; 1152 sdp->sd_tune.gt_statfs_quantum = 30; 1153 } 1154 1155 error = init_names(sdp, silent); 1156 if (error) 1157 goto fail_free; 1158 1159 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name); 1160 1161 error = gfs2_sys_fs_add(sdp); 1162 if (error) 1163 goto fail_free; 1164 1165 gfs2_create_debugfs_file(sdp); 1166 1167 error = gfs2_lm_mount(sdp, silent); 1168 if (error) 1169 goto fail_debug; 1170 1171 error = init_locking(sdp, &mount_gh, DO); 1172 if (error) 1173 goto fail_lm; 1174 1175 error = init_sb(sdp, silent); 1176 if (error) 1177 goto fail_locking; 1178 1179 /* Turn rgrplvb on by default if fs format is recent enough */ 1180 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801) 1181 sdp->sd_args.ar_rgrplvb = 1; 1182 1183 error = wait_on_journal(sdp); 1184 if (error) 1185 goto fail_sb; 1186 1187 /* 1188 * If user space has failed to join the cluster or some similar 1189 * failure has occurred, then the journal id will contain a 1190 * negative (error) number. This will then be returned to the 1191 * caller (of the mount syscall). We do this even for spectator 1192 * mounts (which just write a jid of 0 to indicate "ok" even though 1193 * the jid is unused in the spectator case) 1194 */ 1195 if (sdp->sd_lockstruct.ls_jid < 0) { 1196 error = sdp->sd_lockstruct.ls_jid; 1197 sdp->sd_lockstruct.ls_jid = 0; 1198 goto fail_sb; 1199 } 1200 1201 if (sdp->sd_args.ar_spectator) 1202 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s", 1203 sdp->sd_table_name); 1204 else 1205 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u", 1206 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid); 1207 1208 error = init_inodes(sdp, DO); 1209 if (error) 1210 goto fail_sb; 1211 1212 error = init_per_node(sdp, DO); 1213 if (error) 1214 goto fail_inodes; 1215 1216 error = gfs2_statfs_init(sdp); 1217 if (error) { 1218 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error); 1219 goto fail_per_node; 1220 } 1221 1222 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1223 if (error) 1224 goto fail_per_node; 1225 1226 if (!sb_rdonly(sb)) 1227 error = gfs2_make_fs_rw(sdp); 1228 1229 gfs2_freeze_unlock(&freeze_gh); 1230 if (error) { 1231 fs_err(sdp, "can't make FS RW: %d\n", error); 1232 goto fail_per_node; 1233 } 1234 gfs2_glock_dq_uninit(&mount_gh); 1235 gfs2_online_uevent(sdp); 1236 return 0; 1237 1238 fail_per_node: 1239 init_per_node(sdp, UNDO); 1240 fail_inodes: 1241 init_inodes(sdp, UNDO); 1242 fail_sb: 1243 if (sdp->sd_root_dir) 1244 dput(sdp->sd_root_dir); 1245 if (sdp->sd_master_dir) 1246 dput(sdp->sd_master_dir); 1247 if (sb->s_root) 1248 dput(sb->s_root); 1249 sb->s_root = NULL; 1250 fail_locking: 1251 init_locking(sdp, &mount_gh, UNDO); 1252 fail_lm: 1253 complete_all(&sdp->sd_journal_ready); 1254 gfs2_gl_hash_clear(sdp); 1255 gfs2_lm_unmount(sdp); 1256 fail_debug: 1257 gfs2_delete_debugfs_file(sdp); 1258 gfs2_sys_fs_del(sdp); 1259 fail_free: 1260 free_sbd(sdp); 1261 sb->s_fs_info = NULL; 1262 return error; 1263 } 1264 1265 /** 1266 * gfs2_get_tree - Get the GFS2 superblock and root directory 1267 * @fc: The filesystem context 1268 * 1269 * Returns: 0 or -errno on error 1270 */ 1271 static int gfs2_get_tree(struct fs_context *fc) 1272 { 1273 struct gfs2_args *args = fc->fs_private; 1274 struct gfs2_sbd *sdp; 1275 int error; 1276 1277 error = get_tree_bdev(fc, gfs2_fill_super); 1278 if (error) 1279 return error; 1280 1281 sdp = fc->root->d_sb->s_fs_info; 1282 dput(fc->root); 1283 if (args->ar_meta) 1284 fc->root = dget(sdp->sd_master_dir); 1285 else 1286 fc->root = dget(sdp->sd_root_dir); 1287 return 0; 1288 } 1289 1290 static void gfs2_fc_free(struct fs_context *fc) 1291 { 1292 struct gfs2_args *args = fc->fs_private; 1293 1294 kfree(args); 1295 } 1296 1297 enum gfs2_param { 1298 Opt_lockproto, 1299 Opt_locktable, 1300 Opt_hostdata, 1301 Opt_spectator, 1302 Opt_ignore_local_fs, 1303 Opt_localflocks, 1304 Opt_localcaching, 1305 Opt_debug, 1306 Opt_upgrade, 1307 Opt_acl, 1308 Opt_quota, 1309 Opt_quota_flag, 1310 Opt_suiddir, 1311 Opt_data, 1312 Opt_meta, 1313 Opt_discard, 1314 Opt_commit, 1315 Opt_errors, 1316 Opt_statfs_quantum, 1317 Opt_statfs_percent, 1318 Opt_quota_quantum, 1319 Opt_barrier, 1320 Opt_rgrplvb, 1321 Opt_loccookie, 1322 }; 1323 1324 static const struct constant_table gfs2_param_quota[] = { 1325 {"off", GFS2_QUOTA_OFF}, 1326 {"account", GFS2_QUOTA_ACCOUNT}, 1327 {"on", GFS2_QUOTA_ON}, 1328 {} 1329 }; 1330 1331 enum opt_data { 1332 Opt_data_writeback = GFS2_DATA_WRITEBACK, 1333 Opt_data_ordered = GFS2_DATA_ORDERED, 1334 }; 1335 1336 static const struct constant_table gfs2_param_data[] = { 1337 {"writeback", Opt_data_writeback }, 1338 {"ordered", Opt_data_ordered }, 1339 {} 1340 }; 1341 1342 enum opt_errors { 1343 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW, 1344 Opt_errors_panic = GFS2_ERRORS_PANIC, 1345 }; 1346 1347 static const struct constant_table gfs2_param_errors[] = { 1348 {"withdraw", Opt_errors_withdraw }, 1349 {"panic", Opt_errors_panic }, 1350 {} 1351 }; 1352 1353 static const struct fs_parameter_spec gfs2_fs_parameters[] = { 1354 fsparam_string ("lockproto", Opt_lockproto), 1355 fsparam_string ("locktable", Opt_locktable), 1356 fsparam_string ("hostdata", Opt_hostdata), 1357 fsparam_flag ("spectator", Opt_spectator), 1358 fsparam_flag ("norecovery", Opt_spectator), 1359 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs), 1360 fsparam_flag ("localflocks", Opt_localflocks), 1361 fsparam_flag ("localcaching", Opt_localcaching), 1362 fsparam_flag_no("debug", Opt_debug), 1363 fsparam_flag ("upgrade", Opt_upgrade), 1364 fsparam_flag_no("acl", Opt_acl), 1365 fsparam_flag_no("suiddir", Opt_suiddir), 1366 fsparam_enum ("data", Opt_data, gfs2_param_data), 1367 fsparam_flag ("meta", Opt_meta), 1368 fsparam_flag_no("discard", Opt_discard), 1369 fsparam_s32 ("commit", Opt_commit), 1370 fsparam_enum ("errors", Opt_errors, gfs2_param_errors), 1371 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum), 1372 fsparam_s32 ("statfs_percent", Opt_statfs_percent), 1373 fsparam_s32 ("quota_quantum", Opt_quota_quantum), 1374 fsparam_flag_no("barrier", Opt_barrier), 1375 fsparam_flag_no("rgrplvb", Opt_rgrplvb), 1376 fsparam_flag_no("loccookie", Opt_loccookie), 1377 /* quota can be a flag or an enum so it gets special treatment */ 1378 fsparam_flag_no("quota", Opt_quota_flag), 1379 fsparam_enum("quota", Opt_quota, gfs2_param_quota), 1380 {} 1381 }; 1382 1383 /* Parse a single mount parameter */ 1384 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param) 1385 { 1386 struct gfs2_args *args = fc->fs_private; 1387 struct fs_parse_result result; 1388 int o; 1389 1390 o = fs_parse(fc, gfs2_fs_parameters, param, &result); 1391 if (o < 0) 1392 return o; 1393 1394 switch (o) { 1395 case Opt_lockproto: 1396 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN); 1397 break; 1398 case Opt_locktable: 1399 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN); 1400 break; 1401 case Opt_hostdata: 1402 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN); 1403 break; 1404 case Opt_spectator: 1405 args->ar_spectator = 1; 1406 break; 1407 case Opt_ignore_local_fs: 1408 /* Retained for backwards compat only */ 1409 break; 1410 case Opt_localflocks: 1411 args->ar_localflocks = 1; 1412 break; 1413 case Opt_localcaching: 1414 /* Retained for backwards compat only */ 1415 break; 1416 case Opt_debug: 1417 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC) 1418 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1419 args->ar_debug = result.boolean; 1420 break; 1421 case Opt_upgrade: 1422 /* Retained for backwards compat only */ 1423 break; 1424 case Opt_acl: 1425 args->ar_posix_acl = result.boolean; 1426 break; 1427 case Opt_quota_flag: 1428 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON; 1429 break; 1430 case Opt_quota: 1431 args->ar_quota = result.int_32; 1432 break; 1433 case Opt_suiddir: 1434 args->ar_suiddir = result.boolean; 1435 break; 1436 case Opt_data: 1437 /* The uint_32 result maps directly to GFS2_DATA_* */ 1438 args->ar_data = result.uint_32; 1439 break; 1440 case Opt_meta: 1441 args->ar_meta = 1; 1442 break; 1443 case Opt_discard: 1444 args->ar_discard = result.boolean; 1445 break; 1446 case Opt_commit: 1447 if (result.int_32 <= 0) 1448 return invalfc(fc, "commit mount option requires a positive numeric argument"); 1449 args->ar_commit = result.int_32; 1450 break; 1451 case Opt_statfs_quantum: 1452 if (result.int_32 < 0) 1453 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument"); 1454 args->ar_statfs_quantum = result.int_32; 1455 break; 1456 case Opt_quota_quantum: 1457 if (result.int_32 <= 0) 1458 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument"); 1459 args->ar_quota_quantum = result.int_32; 1460 break; 1461 case Opt_statfs_percent: 1462 if (result.int_32 < 0 || result.int_32 > 100) 1463 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100"); 1464 args->ar_statfs_percent = result.int_32; 1465 break; 1466 case Opt_errors: 1467 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC) 1468 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1469 args->ar_errors = result.uint_32; 1470 break; 1471 case Opt_barrier: 1472 args->ar_nobarrier = result.boolean; 1473 break; 1474 case Opt_rgrplvb: 1475 args->ar_rgrplvb = result.boolean; 1476 args->ar_got_rgrplvb = 1; 1477 break; 1478 case Opt_loccookie: 1479 args->ar_loccookie = result.boolean; 1480 break; 1481 default: 1482 return invalfc(fc, "invalid mount option: %s", param->key); 1483 } 1484 return 0; 1485 } 1486 1487 static int gfs2_reconfigure(struct fs_context *fc) 1488 { 1489 struct super_block *sb = fc->root->d_sb; 1490 struct gfs2_sbd *sdp = sb->s_fs_info; 1491 struct gfs2_args *oldargs = &sdp->sd_args; 1492 struct gfs2_args *newargs = fc->fs_private; 1493 struct gfs2_tune *gt = &sdp->sd_tune; 1494 int error = 0; 1495 1496 sync_filesystem(sb); 1497 1498 spin_lock(>->gt_spin); 1499 oldargs->ar_commit = gt->gt_logd_secs; 1500 oldargs->ar_quota_quantum = gt->gt_quota_quantum; 1501 if (gt->gt_statfs_slow) 1502 oldargs->ar_statfs_quantum = 0; 1503 else 1504 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum; 1505 spin_unlock(>->gt_spin); 1506 1507 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) { 1508 errorfc(fc, "reconfiguration of locking protocol not allowed"); 1509 return -EINVAL; 1510 } 1511 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) { 1512 errorfc(fc, "reconfiguration of lock table not allowed"); 1513 return -EINVAL; 1514 } 1515 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) { 1516 errorfc(fc, "reconfiguration of host data not allowed"); 1517 return -EINVAL; 1518 } 1519 if (newargs->ar_spectator != oldargs->ar_spectator) { 1520 errorfc(fc, "reconfiguration of spectator mode not allowed"); 1521 return -EINVAL; 1522 } 1523 if (newargs->ar_localflocks != oldargs->ar_localflocks) { 1524 errorfc(fc, "reconfiguration of localflocks not allowed"); 1525 return -EINVAL; 1526 } 1527 if (newargs->ar_meta != oldargs->ar_meta) { 1528 errorfc(fc, "switching between gfs2 and gfs2meta not allowed"); 1529 return -EINVAL; 1530 } 1531 if (oldargs->ar_spectator) 1532 fc->sb_flags |= SB_RDONLY; 1533 1534 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) { 1535 struct gfs2_holder freeze_gh; 1536 1537 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1538 if (error) 1539 return -EINVAL; 1540 1541 if (fc->sb_flags & SB_RDONLY) { 1542 gfs2_make_fs_ro(sdp); 1543 } else { 1544 error = gfs2_make_fs_rw(sdp); 1545 if (error) 1546 errorfc(fc, "unable to remount read-write"); 1547 } 1548 gfs2_freeze_unlock(&freeze_gh); 1549 } 1550 sdp->sd_args = *newargs; 1551 1552 if (sdp->sd_args.ar_posix_acl) 1553 sb->s_flags |= SB_POSIXACL; 1554 else 1555 sb->s_flags &= ~SB_POSIXACL; 1556 if (sdp->sd_args.ar_nobarrier) 1557 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1558 else 1559 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1560 spin_lock(>->gt_spin); 1561 gt->gt_logd_secs = newargs->ar_commit; 1562 gt->gt_quota_quantum = newargs->ar_quota_quantum; 1563 if (newargs->ar_statfs_quantum) { 1564 gt->gt_statfs_slow = 0; 1565 gt->gt_statfs_quantum = newargs->ar_statfs_quantum; 1566 } 1567 else { 1568 gt->gt_statfs_slow = 1; 1569 gt->gt_statfs_quantum = 30; 1570 } 1571 spin_unlock(>->gt_spin); 1572 1573 gfs2_online_uevent(sdp); 1574 return error; 1575 } 1576 1577 static const struct fs_context_operations gfs2_context_ops = { 1578 .free = gfs2_fc_free, 1579 .parse_param = gfs2_parse_param, 1580 .get_tree = gfs2_get_tree, 1581 .reconfigure = gfs2_reconfigure, 1582 }; 1583 1584 /* Set up the filesystem mount context */ 1585 static int gfs2_init_fs_context(struct fs_context *fc) 1586 { 1587 struct gfs2_args *args; 1588 1589 args = kmalloc(sizeof(*args), GFP_KERNEL); 1590 if (args == NULL) 1591 return -ENOMEM; 1592 1593 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 1594 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info; 1595 1596 *args = sdp->sd_args; 1597 } else { 1598 memset(args, 0, sizeof(*args)); 1599 args->ar_quota = GFS2_QUOTA_DEFAULT; 1600 args->ar_data = GFS2_DATA_DEFAULT; 1601 args->ar_commit = 30; 1602 args->ar_statfs_quantum = 30; 1603 args->ar_quota_quantum = 60; 1604 args->ar_errors = GFS2_ERRORS_DEFAULT; 1605 } 1606 fc->fs_private = args; 1607 fc->ops = &gfs2_context_ops; 1608 return 0; 1609 } 1610 1611 static int set_meta_super(struct super_block *s, struct fs_context *fc) 1612 { 1613 return -EINVAL; 1614 } 1615 1616 static int test_meta_super(struct super_block *s, struct fs_context *fc) 1617 { 1618 return (fc->sget_key == s->s_bdev); 1619 } 1620 1621 static int gfs2_meta_get_tree(struct fs_context *fc) 1622 { 1623 struct super_block *s; 1624 struct gfs2_sbd *sdp; 1625 struct path path; 1626 int error; 1627 1628 if (!fc->source || !*fc->source) 1629 return -EINVAL; 1630 1631 error = kern_path(fc->source, LOOKUP_FOLLOW, &path); 1632 if (error) { 1633 pr_warn("path_lookup on %s returned error %d\n", 1634 fc->source, error); 1635 return error; 1636 } 1637 fc->fs_type = &gfs2_fs_type; 1638 fc->sget_key = path.dentry->d_sb->s_bdev; 1639 s = sget_fc(fc, test_meta_super, set_meta_super); 1640 path_put(&path); 1641 if (IS_ERR(s)) { 1642 pr_warn("gfs2 mount does not exist\n"); 1643 return PTR_ERR(s); 1644 } 1645 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { 1646 deactivate_locked_super(s); 1647 return -EBUSY; 1648 } 1649 sdp = s->s_fs_info; 1650 fc->root = dget(sdp->sd_master_dir); 1651 return 0; 1652 } 1653 1654 static const struct fs_context_operations gfs2_meta_context_ops = { 1655 .free = gfs2_fc_free, 1656 .get_tree = gfs2_meta_get_tree, 1657 }; 1658 1659 static int gfs2_meta_init_fs_context(struct fs_context *fc) 1660 { 1661 int ret = gfs2_init_fs_context(fc); 1662 1663 if (ret) 1664 return ret; 1665 1666 fc->ops = &gfs2_meta_context_ops; 1667 return 0; 1668 } 1669 1670 static void gfs2_kill_sb(struct super_block *sb) 1671 { 1672 struct gfs2_sbd *sdp = sb->s_fs_info; 1673 1674 if (sdp == NULL) { 1675 kill_block_super(sb); 1676 return; 1677 } 1678 1679 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB); 1680 dput(sdp->sd_root_dir); 1681 dput(sdp->sd_master_dir); 1682 sdp->sd_root_dir = NULL; 1683 sdp->sd_master_dir = NULL; 1684 shrink_dcache_sb(sb); 1685 kill_block_super(sb); 1686 } 1687 1688 struct file_system_type gfs2_fs_type = { 1689 .name = "gfs2", 1690 .fs_flags = FS_REQUIRES_DEV, 1691 .init_fs_context = gfs2_init_fs_context, 1692 .parameters = gfs2_fs_parameters, 1693 .kill_sb = gfs2_kill_sb, 1694 .owner = THIS_MODULE, 1695 }; 1696 MODULE_ALIAS_FS("gfs2"); 1697 1698 struct file_system_type gfs2meta_fs_type = { 1699 .name = "gfs2meta", 1700 .fs_flags = FS_REQUIRES_DEV, 1701 .init_fs_context = gfs2_meta_init_fs_context, 1702 .owner = THIS_MODULE, 1703 }; 1704 MODULE_ALIAS_FS("gfs2meta"); 1705